2 * (c) 2001 Fabrice Bellard
3 * 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * DCT test (c) 2001 Fabrice Bellard
25 * Started from sample code by Juan J. Sierralta P.
37 #include "libavutil/cpu.h"
38 #include "libavutil/common.h"
39 #include "libavutil/lfg.h"
40 #include "libavutil/time.h"
44 #include "simple_idct.h"
46 #include "aandcttab.h"
53 void (*func)(int16_t *block);
54 enum idct_permutation_type perm_type;
59 static const struct algo fdct_tab[] = {
60 { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE },
61 { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE },
62 { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE },
64 { "FAAN", ff_faandct, FF_IDCT_PERM_NONE },
65 #endif /* CONFIG_FAANDCT */
68 static const struct algo idct_tab[] = {
69 { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE },
70 { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 },
71 { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE },
73 { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE },
74 #endif /* CONFIG_FAANIDCT */
75 #if CONFIG_MPEG4_DECODER
76 { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 },
77 #endif /* CONFIG_MPEG4_DECODER */
81 #include "arm/dct-test.c"
83 #include "ppc/dct-test.c"
85 #include "x86/dct-test.c"
87 static const struct algo fdct_tab_arch[] = { { 0 } };
88 static const struct algo idct_tab_arch[] = { { 0 } };
91 #define AANSCALE_BITS 12
94 #define NB_ITS_SPEED 50000
96 DECLARE_ALIGNED(16, static int16_t, block)[64];
97 DECLARE_ALIGNED(8, static int16_t, block1)[64];
99 static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng)
103 memset(block, 0, 64 * sizeof(*block));
107 for (i = 0; i < 64; i++)
108 block[i] = (av_lfg_get(prng) % 512) - 256;
111 for (i = 0; i < 64; i++)
116 j = av_lfg_get(prng) % 10 + 1;
117 for (i = 0; i < j; i++)
118 block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256;
121 block[ 0] = av_lfg_get(prng) % 4096 - 2048;
122 block[63] = (block[0] & 1) ^ 1;
127 static void permute(int16_t dst[64], const int16_t src[64],
128 enum idct_permutation_type perm_type)
133 if (permute_x86(dst, src, perm_type))
138 case FF_IDCT_PERM_LIBMPEG2:
139 for (i = 0; i < 64; i++)
140 dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i];
142 case FF_IDCT_PERM_PARTTRANS:
143 for (i = 0; i < 64; i++)
144 dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i];
147 for (i = 0; i < 64; i++)
153 static int dct_error(const struct algo *dct, int test, int is_idct, int speed)
155 void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct;
158 int64_t err2, ti, ti1, it1, err_sum = 0;
159 int64_t sysErr[64], sysErrMax = 0;
161 int blockSumErrMax = 0, blockSumErr;
166 av_lfg_init(&prng, 1);
170 for (i = 0; i < 64; i++)
172 for (it = 0; it < NB_ITS; it++) {
173 init_block(block1, test, is_idct, &prng);
174 permute(block, block1, dct->perm_type);
179 if (!strcmp(dct->name, "IJG-AAN-INT")) {
180 for (i = 0; i < 64; i++) {
181 scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i];
182 block[i] = (block[i] * scale) >> AANSCALE_BITS;
189 for (i = 0; i < 64; i++) {
190 int err = block[i] - block1[i];
196 sysErr[i] += block[i] - block1[i];
198 if (abs(block[i]) > maxout)
199 maxout = abs(block[i]);
201 if (blockSumErrMax < blockSumErr)
202 blockSumErrMax = blockSumErr;
204 for (i = 0; i < 64; i++)
205 sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i]));
207 for (i = 0; i < 64; i++) {
210 printf("%7d ", (int) sysErr[i]);
214 omse = (double) err2 / NB_ITS / 64;
215 ome = (double) err_sum / NB_ITS / 64;
217 spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015);
219 printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
220 is_idct ? "IDCT" : "DCT", dct->name, err_inf,
221 omse, ome, (double) sysErrMax / NB_ITS,
222 maxout, blockSumErrMax);
224 if (spec_err && !dct->nonspec)
231 init_block(block, test, is_idct, &prng);
232 permute(block1, block, dct->perm_type);
234 ti = av_gettime_relative();
237 for (it = 0; it < NB_ITS_SPEED; it++) {
238 memcpy(block, block1, sizeof(block));
242 ti1 = av_gettime_relative() - ti;
243 } while (ti1 < 1000000);
246 printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name,
247 (double) it1 * 1000.0 / (double) ti1);
252 DECLARE_ALIGNED(8, static uint8_t, img_dest)[64];
253 DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64];
255 static void idct248_ref(uint8_t *dest, int linesize, int16_t *block)
258 static double c8[8][8];
259 static double c4[4][4];
260 double block1[64], block2[64], block3[64];
267 for (i = 0; i < 8; i++) {
269 for (j = 0; j < 8; j++) {
270 s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0);
271 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
272 sum += c8[i][j] * c8[i][j];
276 for (i = 0; i < 4; i++) {
278 for (j = 0; j < 4; j++) {
279 s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0);
280 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
281 sum += c4[i][j] * c4[i][j];
288 for (i = 0; i < 4; i++) {
289 for (j = 0; j < 8; j++) {
290 block1[8 * (2 * i) + j] =
291 (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s;
292 block1[8 * (2 * i + 1) + j] =
293 (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s;
298 for (i = 0; i < 8; i++) {
299 for (j = 0; j < 8; j++) {
301 for (k = 0; k < 8; k++)
302 sum += c8[k][j] * block1[8 * i + k];
303 block2[8 * i + j] = sum;
308 for (i = 0; i < 8; i++) {
309 for (j = 0; j < 4; j++) {
312 for (k = 0; k < 4; k++)
313 sum += c4[k][j] * block2[8 * (2 * k) + i];
314 block3[8 * (2 * j) + i] = sum;
318 for (k = 0; k < 4; k++)
319 sum += c4[k][j] * block2[8 * (2 * k + 1) + i];
320 block3[8 * (2 * j + 1) + i] = sum;
324 /* clamp and store the result */
325 for (i = 0; i < 8; i++) {
326 for (j = 0; j < 8; j++) {
327 v = block3[8 * i + j];
329 else if (v > 255) v = 255;
330 dest[i * linesize + j] = (int) rint(v);
335 static void idct248_error(const char *name,
336 void (*idct248_put)(uint8_t *dest, int line_size,
340 int it, i, it1, ti, ti1, err_max, v;
343 av_lfg_init(&prng, 1);
345 /* just one test to see if code is correct (precision is less
348 for (it = 0; it < NB_ITS; it++) {
349 /* XXX: use forward transform to generate values */
350 for (i = 0; i < 64; i++)
351 block1[i] = av_lfg_get(&prng) % 256 - 128;
354 for (i = 0; i < 64; i++)
355 block[i] = block1[i];
356 idct248_ref(img_dest1, 8, block);
358 for (i = 0; i < 64; i++)
359 block[i] = block1[i];
360 idct248_put(img_dest, 8, block);
362 for (i = 0; i < 64; i++) {
363 v = abs((int) img_dest[i] - (int) img_dest1[i]);
365 printf("%d %d\n", img_dest[i], img_dest1[i]);
370 printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max);
375 ti = av_gettime_relative();
378 for (it = 0; it < NB_ITS_SPEED; it++) {
379 for (i = 0; i < 64; i++)
380 block[i] = block1[i];
381 idct248_put(img_dest, 8, block);
384 ti1 = av_gettime_relative() - ti;
385 } while (ti1 < 1000000);
388 printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name,
389 (double) it1 * 1000.0 / (double) ti1);
392 static void help(void)
394 printf("dct-test [-i] [<test-number>]\n"
395 "test-number 0 -> test with random matrixes\n"
396 " 1 -> test with random sparse matrixes\n"
397 " 2 -> do 3. test from mpeg4 std\n"
398 "-i test IDCT implementations\n"
399 "-4 test IDCT248 implementations\n"
404 #include "compat/getopt.c"
407 int main(int argc, char **argv)
409 int test_idct = 0, test_248_dct = 0;
418 c = getopt(argc, argv, "ih4t");
439 test = atoi(argv[optind]);
441 printf("Libav DCT/IDCT test\n");
444 idct248_error("SIMPLE-C", ff_simple_idct248_put, speed);
446 const int cpu_flags = av_get_cpu_flags();
448 for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++)
449 err |= dct_error(&idct_tab[i], test, test_idct, speed);
451 for (i = 0; idct_tab_arch[i].name; i++)
452 if (!(~cpu_flags & idct_tab_arch[i].cpu_flag))
453 err |= dct_error(&idct_tab_arch[i], test, test_idct, speed);
457 for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++)
458 err |= dct_error(&fdct_tab[i], test, test_idct, speed);
460 for (i = 0; fdct_tab_arch[i].name; i++)
461 if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag))
462 err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed);
464 #endif /* CONFIG_FDCTDSP */
468 printf("Error: %d.\n", err);